• Barnett, T. P., M. Latif, N. Graham, M. Flugel, S. Pazan, and W. White, 1993: ENSO and ENSO-related predictability. Part I: Prediction of equatorial Pacific sea surface temperature with a hybrid coupled ocean–atmosphere model. J. Climate,6, 1545–1566.

  • Barnston, A. G., and C. F. Ropelewski, 1992: Prediction of ENSO episodes using canonical correlation analysis. J. Climate,5, 1316–1345.

  • Cane, M. A., and S. E. Zebiak, 1985: A theory for El Niño and the Southern Oscillation. Science,228, 1085–1087.

  • ——, and ——, 1987: Predictions of El Niño events using a physical model. Atmospheric and Oceanic Variability, H. Cattle, Ed., Royal Meteorological Society Press, 153–182.

  • ——, ——, and S. C. Dolan, 1986: Experimental forecasts of El Niño. Nature,321, 827–832.

  • Chen, D., S. E. Zebiak, A. J. Busalacchi, and M. A. Cane, 1995: An improved procedure for El Niño forecasting. Science,22, 1699–1702.

  • Davey, M. K., S. Ineson, and M. A. Balmaseda, 1994: Simulation and hindcast of tropical Pacific Ocean interannual variability. Tellus,46A, 433–447.

  • DeWitt, D. G., 1996: The effect of the cumulus convection on the climate of the COLA general circulation model. COLA Tech. Rep. 27, 69 pp. [Available from COLA, 4041 Powder Mill Road, Suite 302, Calverton, MD 20705.].

  • DeWitte, B., and C. Perigaud, 1996: El Niño–La Niña events simulated with Cane and Zebiak’s model and observed with satellite or in situ data. Part II: Model forced with observations. J. Climate,9, 1188–1207.

  • Goldenberg, S. B., and J. J. O’Brien, 1981: Time and space variability of the tropical Pacific wind stress. Mon. Wea. Rev.,109, 1190–1207.

  • Halpert, M. S., and C. F. Ropelewski, 1992: Surface temperature patterns associated with the Southern Oscillation. J. Climate,5, 577–593.

  • Harshvardhan, R. Davis, D. A. Randall, and T. G. Corsetti, 1987: A fast radiation parameterization for general circulation models. J. Geophys. Res.,92, 1009–1016.

  • Huang, B., and E. K. Schneider, 1995: The response of an ocean general circulation model to surface wind stress produced by an atmospheric general circulation model. Mon. Wea. Rev.,123, 3059–3085.

  • ——, and J. Shukla, 1997: An examination of the AGCM simulated surface wind stress and low-level winds over the tropical Pacific Ocean. Mon. Wea. Rev,125, 985–998.

  • Jenkins, G. M., and D. G. Watts, 1968: Spectral Analysis and Its Applications. Holden-Day, 525 pp.

  • Ji, M., A. Kumar, and A. Leetmaa, 1994a: An experimental coupled forecast system at the National Meteorological Center: Some early results. Tellus,46A, 398–418.

  • ——, ——, and ——, 1994b: A multi-season climate forecast system at the National Meteorological Center. Bull. Amer. Meteor. Soc.,75, 569–577.

  • ——, A. Leetmaa, and V. E. Kousky, 1996: Coupled model predictions of ENSO during the 1980s and the 1990s at the National Centers for Environmental Prediction. J. Climate,9, 3105–3120.

  • Kinter, J. L., III, J. Shukla, L. Marx, and E. K. Schneider, 1988: A simulation of winter and summer circulations with the NMC global spectral model. J. Atmos. Sci.,45, 2486–2522.

  • Kirtman, B. P., 1997: Oceanic Rossby wave dynamics and the ENSO period in a coupled model. J. Climate,10, 1690–1704.

  • ——, and E. K. Schneider, 1996: Model based estimates of equatorial Pacific wind stress. J. Climate,9, 1078–1091.

  • ——, and D. G. DeWitt, 1997: Comparison of atmospheric model wind stress with three different convective parameterizations: Sensitivity of tropical Pacific Ocean model simulations. Mon. Wea. Rev.,125, 1231–1250.

  • ——, A. Vernekar, D. DeWitt, and J. Zhou, 1993: Impact of orographic gravity wave drag on extended range forecasts with the COLA-GCM. Atmosfera,6, 3–24.

  • ——, J. Shukla, B. Huang, Z. Zhu, and E. K. Schneider, 1997: Multiseasonal predictions with a coupled tropical ocean–global atmosphere system. Mon. Wea. Rev.,125, 789–808.

  • Kleeman, R., 1993: On the dependence of hindcast skill on ocean thermodynamics in a coupled ocean–atmosphere model. J. Climate,6, 2012–2033.

  • Kuo, H. L., 1965: On the formation and intensification of tropical cyclones through latent heat release by cumulus convection. J. Atmos. Sci.,22, 40–63.

  • Lacis, A. A., and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth’s atmosphere. J. Atmos. Sci.,31, 118–133.

  • Latif, M., and A. Villwock, 1990: Interannual variability as simulated in coupled ocean–atmosphere models. J. Mar. Syst.,1, 51–60.

  • ——, A. Sterl, E. Maier-Reimer, and M. M. Junge, 1993a: Climate variability in a coupled GCM. Part I: The tropical Pacific. J. Climate,6, 5–21.

  • ——, ——, ——, and ——, 1993b: Structure and predictability of the El Niño/Southern Oscillation phenomena in a coupled ocean–atmosphere general circulation model. J. Climate,6, 700–708.

  • ——, T. P. Barnett, M. A. Cane, M. Flugel, N. E. Graham, H. von Storch, J.-S. Xu, and S. E. Zebiak, 1994: A review of ENSO prediction studies. Climate Dyn.,9, 167–179.

  • Leetmaa, A., and M. Ji, 1989: Operational hindcasting of the tropical Pacific. Dyn. Atmos. Oceans,13, 465–490.

  • Mellor, G. L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys.,20, 851–875.

  • Miyakoda, K., and J. Sirutis, 1977: Comparative integrations of global spectral models with various parameterized processes of sub-grid scale vertical transports. Beitr. Phys. Atmos.,50, 445–48.

  • Neelin, J. D., 1990: A hybrid coupled general circulation model for El Niño studies. J. Atmos. Sci.,47, 674–693.

  • Pacanowski, R. C., K. Dixon, and A. Rosati, 1993: The GFDL modular ocean model users guide, version 1.0. GFDL Ocean Group Tech. Rep. 2, 77 pp. [Available from NOAA/GFDL, Princeton University, Princeton, NJ 08542.].

  • Parker, D. E., M. Jackson, and E. B. Horton, 1995: The GISST sea surface temperature and sea ice climatology. Hadley Centre Climate Research Tech. Note 63, 49 pp. [Available from Hadley Centre Meteorological Office, London Road, Bracknell, Berkshire RG12 2SY, United Kingdom.].

  • Penland, C., and T. Magorian, 1993: Prediction of Niño 3 sea surface temperatures using linear inverse modeling. J. Climate,6, 1067–1076.

  • Philander, S. G. H., R. C. Pacanowski, N. C. Lau, and M. J. Nath, 1992: Simulation of ENSO with a global atmospheric GCM coupled to a high-resolution, tropical Pacific Ocean GCM. J. Climate,5, 308–329.

  • Robertson, A. W., C.-C. Ma, M. Ghil, and C. R. Mechoso, 1995: Simulation of the tropical Pacific climate with a coupled ocean–atmosphere general circulation model. Part II: Interannual variability. J. Climate,8, 1199–1216.

  • Reynolds, R. W., 1988: A real time global sea surface temperature analysis. J. Climate,1, 75–86.

  • Schneider, E. K., and J. L. Kinter III, 1994: An examination of internally generated variability in long climate simulations. Climate Dyn.,10, 181–204.

  • ——, Z. Zhu, B. Huang, B. Giese, B. P. Kirtman, and J. Shukla, 1997: ENSO variability in a coupled general circulation model. Mon. Wea. Rev., in press.

  • Spencer, R. W., 1993: Global oceanic precipitation from the MSU during 1979–91 and comparison to other climatologies. J. Climate,6, 1301–1326.

  • Tiedtke, M., 1984: The effect of penetrative cumulus convection on the large-scale flow in a general circulation model. Beitr. Phys. Atmos.,57, 216–239.

  • Trenberth, K. E., W. G. Large, and J. G. Olson, 1990: The mean annual cycle in global ocean wind stress. J. Phys. Oceanogr.,20, 1742–1760.

  • Xue, Y., P. J. Sellers, J. L. Kinter III, and J. Shukla, 1991: A simple biosphere model for global climate studies. J. Climate,4, 345–364.

  • Zebiak, S. E., and M. A. Cane, 1987: A model of El Niño and the Southern Oscillation. Mon. Wea. Rev.,115, 2262–2278.

  • ——, and ——, 1991: Natural climate variability in a coupled model. Workshop on Greenhouse-Gas-Induced Climate Change: A Critical Appraisal of Simulations and Observations, M. E. Schlesinger, Ed., Elsevier, 457–470.

  • View in gallery

    Temporal correlation between uncoupled ocean model simulated and observed SSTA. In (a) the wind stress anomaly is determined directly from the AGCM and in (b) the AGCM wind stress is empirically derived from the AGCM 850-mb winds. In (c) the wind stress is determined from the FSU analysis. In (d) the wind stress is determined from the ocean model iteration procedure.

  • View in gallery

    NINO3 (5°S–5°N, 150°–90°W) averaged sea surface temperature anomaly for the HCM simulation. In (a) simulation years 1–22 and in (b) simulation years 23–44 are shown.

  • View in gallery

    NINO3 (5°S–5°N, 150°–90°W) averaged sea surface temperature anomaly for the observations. In (a) the SSTA for 1949–69 and in (b) the SSTA 1970–91 is shown.

  • View in gallery

    Power spectra of the (a) observed and (b) simulated NINO3 SSTA. The solid curve shows the power spectra, the long-dashed curve shows the power spectra for red noise, and the short-dashed curve shows the red noise power spectra at the 95% confidence level.

  • View in gallery

    Time–longitude cross section of (a) simulated and (b) observed SSTA along the equator. For the HCM, the simulation years 33–44 are shown, and for the observation, 1982–93 is shown.

  • View in gallery

    Time–longitude cross section of (a) simulated and (b) observed zonal wind stress anomaly along the equator. For the HCM, the simulation years 33–44 are shown, and for the observation, 1982–93 is shown.

  • View in gallery

    Time–longitude cross section of (a) simulated and (b) observed precipitation anomaly along the equator. For the HCM, the simulation years 33–44 are shown, and for the observation, 1982–93 is shown.

  • View in gallery

    SSTA first empirical orthogonal function of the (a) HCM, (b) observations, and (c) COLA prediction system. The explained variance is noted in each panel.

  • View in gallery

    NINO3 SSTA, for the observations (solid curve), the uncoupled ocean model before (dot-dashed curve) and after (short-dashed curve) the iteration procedure.

  • View in gallery

    NINO3 SSTA systematic error in the HCM (top panel) and the CZBC (bottom panel) hindcasts. The thick solid curve in the top panel shows the observed annual cycle scaled by a factor of 3. The systematic error of the January (dot–dashed curve), the April (short-dashed curve), the July (long-dashed curve), and the October (thin solid curve) hindcast are shown separately.

  • View in gallery

    Evolution of the NINO3 SSTA observations (thick solid curve) and the hindcasts (short-dashed curve). The top panel shows the HCM hindcasts, the middle panel shows the COLA hindcasts, and the bottom panel show the CZBC hindcasts.

  • View in gallery

    NINO3 SSTA (a) correlation coefficient and (b) root-mean-square error (rmse). The skill of the HCM is shown in the solid curve and the skill of the COLA system is shown in the long-dashed curve. The skill of the CZBC system is shown in the short-dashed curve, and the skill of persistence is shown in the dot–dot-dashed curve.

  • View in gallery

    SSTA correlation of the HCM hindcasts as a function of latitude, longitude, and time in (a), (c), (e), and (g). SSTA correlation of a persistence hindcast as a function of latitude, longitude, and time in (b), (d), (f), and (h).

  • View in gallery

    (a) HCM precipitation anomaly correlation coefficient and (b) persistence precipitation anomaly correlation coefficient for lead time of 5–7 months.

  • View in gallery

    Time–longitude cross section along the equator of the (a) January 1982 hindcast and (b) observed SSTA.

  • View in gallery

    Time–longitude cross section along the equator of the (a) January 1982 hindcast and (b) observed precipitation anomaly.

  • View in gallery

    Time–longitude cross section along the equator of the (a) July 1981 hindcast and (b) observed SSTA.

  • View in gallery

    Time–longitude cross section along the equator of the (a) January 1983 hindcast and (b) observed SSTA.

  • View in gallery

    Time–longitude cross section along the equator of the (a) January 1984 hindcast and (b) observed SSTA.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 54 18 0
PDF Downloads 30 12 0

ENSO Simulation and Prediction with a Hybrid Coupled Model

View More View Less
  • 1 Center for Ocean–Land–Atmosphere Studies, Institute of Global Environment and Society, Inc., Calverton, Maryland
  • | 2 Lamont-Doherty Earth Observatory of Columbia University, Palisades, New York
Full access

Abstract

A hybrid coupled model (HCM) consisting of a tropical Pacific Ocean and global atmosphere is presented. The ocean component is a linear reduced gravity model of the upper ocean in the tropical Pacific. The atmospheric component is a triangular 30 horizontal resolution global spectral general circulation model with 18 unevenly spaced levels in the vertical. In coupling these component models, an anomaly coupling strategy is employed. A 40-yr simulation was made with HCM and the variability in the tropical Pacific was compared to the observed variability. The HCM produces irregular ENSO events with a broad spectrum of periods between 12 and 48 months. On longer timescales, approximately 48 months, the simulated variability was weaker than the observed and on shorter timescales (approximately 24 months) the simulated variability was too strong. The simulated variability is asymmetric in the sense that the amplitude of the warm events is realistic, but there are no significant cold events.

An ensemble of 60 hindcast predictions was made with the HCM and the skill was compared to other prediction systems. In forecasting sea surface temperature anomalies in the eastern Pacific, the HCM is comparable to the other prediction systems for lead times up to 10 months. The anomaly correlation coefficient for the eastern Pacific SSTA remains above 0.6 for lead times of up to 11 months. Consistent with the 40-yr simulation, hindcasts of cold events have little skill, particularly when compared to hindcasts of warm events. Specific hindcasts also demonstrate that the HCM also has difficulty predicting the transition from warm conditions to normal or cold conditions.

Corresponding author address: Dr. Ben P. Kirtman, Center for Ocean–Land–Atmosphere Studies, Institute of Global Environment and Society, Inc., 4041 Powdermill Road, Suite 302, Calverton, MD 20705-3106.

Email: kirtman@cola.iges.org

Abstract

A hybrid coupled model (HCM) consisting of a tropical Pacific Ocean and global atmosphere is presented. The ocean component is a linear reduced gravity model of the upper ocean in the tropical Pacific. The atmospheric component is a triangular 30 horizontal resolution global spectral general circulation model with 18 unevenly spaced levels in the vertical. In coupling these component models, an anomaly coupling strategy is employed. A 40-yr simulation was made with HCM and the variability in the tropical Pacific was compared to the observed variability. The HCM produces irregular ENSO events with a broad spectrum of periods between 12 and 48 months. On longer timescales, approximately 48 months, the simulated variability was weaker than the observed and on shorter timescales (approximately 24 months) the simulated variability was too strong. The simulated variability is asymmetric in the sense that the amplitude of the warm events is realistic, but there are no significant cold events.

An ensemble of 60 hindcast predictions was made with the HCM and the skill was compared to other prediction systems. In forecasting sea surface temperature anomalies in the eastern Pacific, the HCM is comparable to the other prediction systems for lead times up to 10 months. The anomaly correlation coefficient for the eastern Pacific SSTA remains above 0.6 for lead times of up to 11 months. Consistent with the 40-yr simulation, hindcasts of cold events have little skill, particularly when compared to hindcasts of warm events. Specific hindcasts also demonstrate that the HCM also has difficulty predicting the transition from warm conditions to normal or cold conditions.

Corresponding author address: Dr. Ben P. Kirtman, Center for Ocean–Land–Atmosphere Studies, Institute of Global Environment and Society, Inc., 4041 Powdermill Road, Suite 302, Calverton, MD 20705-3106.

Email: kirtman@cola.iges.org

1. Introduction

It is now well recognized that the El Niño–Southern Oscillation (ENSO) phenomenon results from coupled ocean–atmosphere interactions in the tropical Pacific. Moreover, given the large global climatic impacts associated with ENSO (Halpert and Ropelewski 1992), it is not surprising that there have been a number of coupled modeling efforts designed to understand and predict ENSO variations in the tropical Pacific. Latif et al. (1997, manuscript submitted to J. Geophys. Res.) and Latif et al. (1994) and references therein contain a comprehensive review of ENSO prediction and predictability research.

Several coupled model strategies have been employed in prediction schemes to date including purely dynamical and mixed dynamical–statistical methods.1 The coupled general circulation model (CGCM) approach uses a dynamical ocean general circulation model (OGCM) coupled to an atmospheric general circulation model (AGCM). This CGCM method is used by Kirtman et al. (1997) and K. Miyakoda et al. (1997, manuscript submitted to Mon. Wea. Rev.) for experimental ENSO prediction and by Ji et al. (1996), Ji et al. (1994a and 1994b), and Leetmaa and Ji (1989) for operational ENSO prediction at the National Centers for Environmental Prediction (NCEP). The CGCM method has also been used in ENSO simulation studies (Philander et al. 1992; Latif et al. 1993a,b; Robertson et al. 1995; Schneider et al. 1997).

The intermediate coupled model (ICM) approach uses simplified dynamically based models of the ocean and the atmosphere and has been used for ENSO prediction and simulation (e.g., Cane and Zebiak 1985; Cane et al. 1986; Zebiak and Cane 1987; Cane and Zebiak 1987; Kleeman 1993; Chen et al. 1995). Chen et al. (1995, hereafter CZBC) introduced a coupled initialization procedure into an ICM that substantially improved the hindcast skill in the eastern Pacific and sets a new standard for coupled model ENSO prediction efforts.

The hybrid coupled models (HCM) generally use OGCMs coupled to either a simplified dynamical or statistical atmosphere model (Neelin 1990; Latif and Villwock 1990; Barnett et al. 1993; Davey et al. 1994). The strategy employed here is a variation of the HCM method in which a sophisticated AGCM is coupled to an intermediate level anomaly ocean model of the tropical Pacific. While our strategy differs from the earlier HCMs in that we have chosen to use a simplified ocean instead of a simplified atmosphere, the general hybrid concept of one sophisticated component and one simplified component is similar and we will refer to the coupled model presented here as an HCM.

This HCM consists of the Center for Ocean–Land–Atmosphere Studies (COLA) AGCM coupled to the ocean component of the Zebiak and Cane (1987) coupled model (hereafter referred to as the ZC ocean model). The motivation for the HCM strategy and the choice of this particular ocean model is, in part, based on the CGCM hindcasts presented in Kirtman et al. (1997, hereafter referred to as the COLA prediction system). The COLA prediction system also used the COLA AGCM, but coupled to Geophysical Fluid Dynamics Laboratory (GFDL) OGCM. Because of errors in the AGCM wind stress climatology and the OGCM sea surface temperature (SST) climatology, an anomaly coupling strategy was employed in which the ocean and atmosphere models exchange anomalies superimposed on observed climatologies of surface wind stress and SST. While the anomaly coupling strategy removes the errors in the climatology at the interface of the two models, it cannot alleviate errors in the component models and how these errors affect the predicted anomalies. For example, using the GFDL OGCM, Kirtman and DeWitt (1997) argued that errors in the climatological equatorial thermocline inhibited the remote sea surface temperature anomaly (SSTA) response in the eastern Pacific to anomalous wind stress forcing in the central and western Pacific. The ZC ocean model uses a specified climatology for the entire ocean, not just at the interface of the ocean and atmosphere, so that there is no negative impact on the predicted anomaly due to errors in the ocean model climatology.

Based on their hindcasts, Kirtman et al. (1997) also computed a composite hindcast SSTA and compared it to an observed composite. The most outstanding feature of the hindcast composite was that the meridional scale of the SSTA was substantially smaller than the observed composite. This meridional-scale problem was also observed in uncoupled ocean simulations with the same OGCM (Kirtman and Schneider 1996); however, the problem becomes more severe in the CGCM due to coupled feedbacks between the ocean and atmosphere. Kirtman et al. (1997) speculated that this meridional-scale problem potentially limits the predictability of their coupled prediction system. Moreover, these systematic errors in the simulated SSTA can lead to systematic errors in the simulation of tropical rainfall anomalies and the associated remote atmospheric response. The ZC ocean model and, as a consequence, the HCM presented here suffer much less from this meridional-scale problem.

The best way to address the above problems is to improve the OGCM climatology; unfortunately, this is not necessarily easy to do. Instead, we have selected an ocean model that improves these particular problems, at the possible expense of introducing a different set of errors into the coupled system. In any case, it is of interest to investigate the simulated ENSO variability and hindcast skill when the same AGCM is coupled to different ocean models and when the same ocean model is coupled to different atmospheric components.

The paper briefly describes the component models, the coupling procedure, and some empirical fixes applied to the AGCM wind stress (e.g., Huang and Shukla, 1997). To facilitate comparison, the atmospheric component used here is identical to that used in the COLA prediction system. A 40-yr extended simulation was made with the HCM and is analyzed here. The HCM’s ability to hindcast the period of 1980–94 is analyzed based on an ensemble of 60 hindcast predictions initialized during each season of each year, with initialization procedure following that suggested by Kirtman and Schneider (1996).

2. The component models

The atmosphere and ocean models used in this study are described below. The atmospheric model has been used in coupled model studies (Schneider et al. 1997) and in previous ENSO prediction efforts (Kirtman et al. 1997). The ocean component has also been used for prediction and simulation (Cane et al. 1986; Zebiak and Cane 1987; CZBC) as well as for mechanistic studies of ENSO (e.g., Zebiak and Cane 1991; Kirtman 1997).

a. The COLA AGCM

The atmospheric model used in the experiments described here is the COLA AGCM, and more details can be found in Schneider and Kinter (1994), Xue et al. (1991), and Kinter et al. (1988). The model is a global spectral model with triangular truncation at wavenumber 30. There are 18 unevenly spaced σ-coordinate vertical levels. The model includes the simplified biosphere model over land described in Xue et al. (1991), the parameterization of the solar radiation is after Lacis and Hansen (1974), and the terrestrial radiation follows Harshvardhan et al. (1987). The deep convection scheme incorporates a modified version of the Kuo (1965) scheme and the shallow convection follows Tiedtke (1984). There is a second-order turbulent closure scheme for subgrid-scale exchanges of heat, momentum, and moisture as in Miyakoda and Sirutis (1977) and Mellor and Yamada (1982) at their closure level 2.0. Surface wave drag and the vertical distribution of the wave drag due to vertically propagating gravity waves are parameterized following Kirtman et al. (1993).

When the atmospheric model was forced with observed SST and the resulting model output wind stress was used to force the GFDL OGCM, errors were found in the OGCM simulation of the SSTA and the heat content anomaly that were directly linked to errors in the AGCM wind stress anomaly (Huang and Schneider 1995). These errors in the COLA AGCM wind stress appear to be most severe during boreal spring when the anomaly becomes too weak and confined to the western extreme of the Pacific basin. Huang and Shukla (1997) found that these errors were due to errors within the boundary layer and that the winds at the top of the boundary (approximately 850 mb) can be effectively converted into a surface stress that yields a substantially improved uncoupled ocean simulation and coupled predictions (Kirtman et al. 1997). This empirical correction involves fitting the AGCM 850-mb zonal wind anomaly to an observed zonal wind stress anomaly calculated from The Florida State University (FSU) subjectively analyzed pseudostress (Goldenberg and O’Brien 1981). The procedure for converting the FSU pseudostress into a wind stress follows Trenberth et al. (1990). This empirical correction to the wind stress anomaly is applied in the HCM in precisely the same manner as in the COLA prediction system, and its impact on the uncoupled ZC ocean model simulation is briefly discussed below.

b. The ZC ocean model

The dynamics of the ZC ocean model is described by linear shallow-water equations, which produce thermocline depth anomalies and depth-averaged baroclinic currents. A shallow frictional layer of constant depth (50 m) is embedded to simulate the surface intensification of wind-driven currents. A relatively complete heat budget is carried in the surface layer, including horizontal and vertical advection of temperature by both mean and anomalous currents. The annual cycle is included in the model by the prescribed mean currents, temperature, and thermocline depth. Surface heat fluxes are simplified to a form that acts only to damp the SSTA to zero with an e-folding timescale of 125 days. The ocean model time step is 10 days.

To assess the impact of the empirical 850-mb correction to the AGCM wind stress anomaly discussed above, we have forced the uncoupled ZC ocean model both with and without the empirical correction to the wind stress anomaly. Figures 1a and 1b show the SSTA temporal correlation based on a 30-yr simulation (1964–93) where the correlation is computed over 1970–93 allowing for a 6-yr spinup of the ocean model and, for reference, Fig. 1c shows the correlation when the FSU wind stress has been applied. The AGCM wind stress with and without the empirical correction is calculated from an uncoupled AGCM simulation with observed SST. The global sea ice and sea surface temperature (GISST, Parker et al. 1995) is used as the observed data in the uncoupled AGCM simulation and in calculating the correlation coefficient. The SSTA correlation using the FSU data to force the ZC ocean model (Fig. 1c) and using the uncorrected AGCM wind stress (Fig. 1a) are comparable with correlations greater than 0.4 throughout a substantial region of the central and eastern tropical Pacific. Throughout most of the equatorial Pacific, the empirical correction to the wind stress anomaly improves the simulation of the SSTA giving correlations greater than 0.6 over much of the equatorial central and eastern Pacific. The fact that the empirically corrected AGCM wind stress improves the SSTA simulation over the FSU wind stress is consistent with the results of Kirtman and Schneider (1996) and leads us to believe it will improve the simulation when it is incorporated into the HCM. Finally, for comparison, Fig. 1d shows the correlation after the Kirtman and Schneider (1996) iteration procedure is applied to the ZC ocean model. It is this final simulation that is used as initial conditions in the hindcast experiments to be discussed in section 4.

c. Coupling procedure

In coupling the COLA AGCM to the ZC ocean model, we have employed a procedure similar to the COLA prediction system in that the component models exchange only predicted anomalies. The coupling procedure is as follows. Given an SST field, the AGCM produces a total wind stress field that has been empirically corrected as described above. The AGCM wind stress climatology is subtracted and the wind stress anomaly is passed to the ocean component. The AGCM wind stress climatology is computed with respect to an uncoupled simulation with observed SST. Given a wind stress anomaly, the ZC ocean model produces a predicted SSTA for the tropical Pacific. The SSTA is superimposed on the observed global annually varying SST climatology and is then passed to the AGCM. Since the ocean model time step is 10 days, the atmosphere and ocean components exchange information once every 10 days. The oceanic and atmospheric grids are not the same and the anomalies must be interpolated to the corresponding component model grids. The spatial interpolation is linear and uses routines provided in the Modular Interface for Coupled Air Sea Applications (MICASA; Pacanowski et al. 1993).

3. Extended simulation

In this section, a 40-yr extended simulation of the HCM is presented and compared to available observational data. The extended simulation is a continuation of the hindcast initialized in July 1982. The procedure for initializing the hindcasts is described in the next section. In this section, we focus on the interannual variability in the tropical Pacific simulated by the HCM over a 40-yr period. We have arbitrarily selected a 40-yr period from the observational record for comparison. Given that the comparison is between an extended model simulation and the observed time series, only the character and composite structure of the oscillation can be compared.

Figures 2a,b and 3a,b show the simulated and observed SSTA variability in the NINO3 region (5°S–5°N, 150°–90°W), respectively. The observed SSTA is defined as the deviation from the mean annual cycle calculated over the entire record (1949–90) and the HCM SSTA is the anomaly from the CZ ocean model climatology. Similar to the observed time series, the HCM variability is irregular. There are extended quiescent periods, such as simulation years 29–34. Similar inactive periods can be identified in the observed time series, such as during 1958–63. The simulated and observed time series have nearly the same standard deviation (0.76°C). The thin straight lines denote plus and minus one standard deviation. There are five warm events (years 5, 11, 19, 37, and 42) where the simulated SSTA is greater than one standard deviation for more than a couple of months. There are no significant cold events in the simulation, except perhaps in year 44. In the observed time series, there are five major warm events (1958, 1966, 1973, 1983, and 1987) and five major cold events (1956, 1971, 1974, 1976, and 1988), although the amplitude of the cold events is less than the warm events. The inability of the HCM to simulate substantial cold events can also be seen in the relatively poor hindcast skill of the cold events discussed later in the paper.

In the frequency domain there are some important differences between the simulated and observed NINO3 variability. Figures 4a and 4b show the spectral density of the simulation and the observed data, respectively. The spectral density calculation (for both the time series and red noise) is based on the Fourier method and the confidence interval is determined from a chi-squared table. For additional details the reader is referred to Jenkins and Watts (1968). The dominant observed period is about 42 months, although the spectral density is greater than red noise at the 95% confidence level for periods between 36 and 63 months. The simulated spectral density has a broader spectrum of periods with significantly less power at lower frequencies and somewhat larger variance at higher frequencies, particularly 24 months. The simulation also has power on the annual timescale, which can also be detected in Figs. 2a and 2b.

Time–longitude cross sections of the simulated and observed SSTA, wind stress anomaly, and precipitation anomaly indicate that the HCM variability is similar in character to the observed interannual variability. Figures 5a and 5b show the SSTA along the equator for simulation years 33–44 and the observed SSTA for 1982–93. Both the observed SSTA and the simulated SSTA are dominated by a standing oscillation with the largest amplitude in the eastern Pacific. The simulated SSTA is smoother than the observed SSTA. In the simulated SSTA, there is a clear asymetery between warm and cold periods with cold anomalies only slightly below −1°C and warm anomalies well above 2°C. In contrast, in the observed data, there are substantial regions along the equator where the anomalies are well below −1°C for significant periods of time. The tendency of the HCM to produce weak-amplitude higher-frequency (approximately 12-month period) oscillations can also be seen in simulation years 33–37.

In the same format as Figs. 5a and 5b, Figs. 6a and 6b and Figs. 7a and 7b show time–longitude cross sections along the equator of the zonal wind stress anomaly and the precipitation anomaly, respectively. In both the observed and simulated wind stress anomaly, the variability is largest in the central Pacific. Consistent with the simulated SSTA, the easterly anomalies are relatively weak compared to the westerly anomalies. The structure of the zonal wind stress anomaly is relatively simple when compared to the observed because the 850-mb empirical formulation reduces the zonal wind stress spatial variability and because the ZC ocean model has too little spatial variability in the SSTA.

The observed and simulated precipitation anomalies are also consistent with the observed and the simulated SSTA, respectively. In the observed data (Fig. 7b), the largest amplitude negative rainfall anomalies associated with large cold events occur in the west-central Pacific between 150°W and 180°. During warm events, the positive rainfall anomalies extend into the far eastern Pacific. In the simulation (Fig. 7a), there is relatively weak precipitation variability to the west of 180° and almost no negative rainfall anomalies. To the east of 180°, the simulated precipitation variability is similar to the observed with a fairly large response to the warm SSTA.

The absence of significant cold events in the simulation is most likely due to problems with the ocean model. This conclusion is based on the fact that the COLA prediction system produces stronger cold events in extended integrations and in hindcast experiments. In addition, with the empirical corrections to the wind stress anomalies, the AGCM produces reasonably realistic easterly anomalies when forced by observed SST. On the other hand, the ZC ocean model, when forced with observed (or AGCM wind stress), produces a relatively poor simulation of cold events. This relatively poor simulation of cold events is also seen in the next section.

While the simulation of cold events is degraded by the simpler ocean component, the spatial structure of the SSTA is improved. The differences in the structure of the SSTA in the COLA prediction system and the HCM can be clearly seen in Figs. 8a–c. Figures 8a–c show the first empirical orthogonal function (EOF) of the HCM, the observed, and the COLA prediction system SSTA. The explained variance is also noted in each figure. A 1-2-1 temporal filter has been applied to the SSTA data before the EOF calculation, which tends to increase the explained variance. In calculating the HCM EOF, we have used the last 30 yr of this 40-yr simulation. For the observed SSTA, we used 30 yr (1964–93) of the GISST data. The COLA SSTA EOF is calculated based on sixty 18-month hindcast experiments.

The maximum amplitude of the HCM EOF is in the far eastern Pacific along the coast of South America. There is also a fairly strong zonal gradient in the HCM EOF as the anomaly decays rapidly to the west. The observed maximum occurs somewhat displaced from the eastern boundary and, in comparison with the HCM, has less zonal variation. The maximum in the COLA EOF is distinctly displaced to the west in comparison to either the observed or the HCM EOF. There is also a suggestion that the COLA EOF extends too far to the west. The most striking difference between the HCM and the COLA EOF, however, is in the meridional scale of the anomaly. While the meridional scale of the HCM EOF is not as broad as the observed, it is a marked improvement over the COLA EOF, particularly in the far eastern Pacific.

The problems with the meridional scale of the SSTA in the OGCM is not particularly well understood. The OGCM has a narrow structure that is confined to the immediate vicinity of the upwelling along the equator. Part of the problem is due to the parameterized heat flux that tends to damp the off-equatorial SSTA too much. The climatological meridional SST gradients in the OGCM are weak compared to the observed gradients, particularly in the eastern Pacific in the vicinity of the oceanic ITCZ. These weak gradients in the OGCM also contribute to the rather weak SSTA just off the equator. The simpler ocean model avoids some of these problems by a coarser SST grid, by a strong poleward mean Ekman flux, and by the fact that the SST equation uses the observed horizontal structure of the climatological SST.

4. Prediction experiments

In this section, the results of 60 hindcast predictions with the HCM are compared to predictions made with the COLA and CZBC prediction systems. The procedure for initializing the hindcast is described, the systematic error of the hindcasts is shown, and an overall assessment of the SSTA hindcast skill is presented.

a. Initialization

The initialization of the ZC ocean model for the coupled hindcasts is similar to that used in the COLA prediction system. Motivated by the need to generate better ocean initial conditions without necessarily assimilating subsurface ocean data, Kirtman and Schneider (1996) developed an iterative wind stress initialization procedure. The wind stress is modified to correct the simulated SSTA error. A linear adjustment was applied that was local in both space and time where for each 1 K of SSTA simulation error, the zonal wind stress anomaly was adjusted by 0.1 dyn cm−2. The iteration procedure is to run the uncoupled ZC ocean model with the wind stress from the AGCM forced by observed SST for 30 yr. Once the 30-yr simulation is completed, the AGCM wind stress at each grid point and each month is adjusted based on the simulated SSTA error. The uncoupled ocean model is then integrated for another 30 yr with the modified wind stress. The final product is a better SSTA simulation as measured by the SSTA correlation and root-mean-square error (rmse) and possibly a better zonal wind stress anomaly.

The impact of the iteration procedure on the SSTA in the eastern Pacific in the ZC ocean model can be seen in Fig. 9, where the NINO3 SSTA is plotted for 1982–92. The SSTA for the uncoupled simulation before the iteration is the shown in the dot–dashed curve and the short-dashed curve shows the SSTA after the iteration is applied. The observed NINO3 SSTA is indicated by the solid curve. The largest impact of the iteration is to enhance the amplitude of cold SSTA, particularly in 1984–86 and 1988–89. Although less dramatic, there is some enhancement of the amplitude during warm events. The iteration also introduces some shorter timescale features to the simulation.

Given that the modification to the zonal wind stress anomaly depends on the ocean model simulation, the iteration procedure may be introducing errors into the zonal wind stress that compensate for errors in the ocean model. In other words, the SSTA simulation may be improved, but it may also be for the wrong reasons. As evidenced in Fig. 9, the iteration procedure compensates for a systematic problem in simulating cold events with the ZC ocean model. As a consequence, it is possible that errors have been introduced into the zonal wind stress that appear as temporally and spatially remote errors in the subsurface heat content. Kirtman and Schneider (1996) also considered the simulation of the subsurface heat content along the equator and found that the iteration procedure introduces only very modest changes in the subsurface heat content over the verification period. Moreover, Kirtman et al. (1997) found that the iteration improved their forecasts, particularly for short lead times. Although not shown here, we have also found in controlled hindcast experiments that the iteration procedure improves the skill of the hindcasts, particularly for short lead times.

We have completed 60 hindcasts initialized in January, April, July, and October for the years 1980–94. Each hindcast was run for 18 months. The ocean initial conditions were taken from the uncoupled ocean simulation after the iteration procedure was applied. The atmospheric initial conditions are taken from an uncoupled AGCM simulation with observed SST. To verify the hindcast SSTA for the cases initialized after 1981, we use Reynolds (1988) blended SST analysis. Before 1981, the GISST is used for verification.

b. Systematic error

Before all the hindcasts are analyzed and compared to observational data, the systematic error is removed. The systematic SSTA error for all 15 forecasts initialized in January, for example, is defined as
i1520-0493-125-10-2620-eq1
where, in this example, the summation is over all forecasts initialized in January so that N = 15 and SSTAj is the SSTA of the jth January hindcast and τ is the lead time. Since each hindcast is 18 months in duration, there are 18 months of systematic error. The systematic error is defined in the same way for the hindcasts initialized in April, July, and October. In addition, the systematic error for other fields is defined in the same manner.

Figures 10a and 10b show the SSTA systematic error for the HCM and the CZBC prediction systems, respectively, for the hindcasts initialized in January, April, July, and October. The systematic errors are computed over the same 60 cases for both prediction systems. The thick solid curve in Fig. 10a shows the annual cycle of the observed SST divided by a factor of 3. The evolution of the HCM systematic error for all four initial months is phase locked with the annual cycle with maximum positive error in boreal spring and maximum negative error in boreal fall. This phase locking of the systematic error with the annual cycle is likely due to the fact that the HCM and the uncoupled AGCM have slightly different climatological annual cycles. The AGCM wind stress annual cycle, which is a priori specified in the anomaly coupling strategy of the HCM, is computed from an extended integration of the AGCM with observed SST. The HCM, however, uses climatological SST globally plus simulated SSTA with in the ocean model domain and, therefore, has a slightly different mean annual cycle. These differences produce a weak annual cycle in the HCM wind stress anomaly, which forces a systematic error in the HCM SSTA that is phase locked to the observed SST annual cycle.

The character of the CZBC systematic error (Fig. 10b) is significantly different than the HCM. In this case, the systematic error is positive definite for all lead times and all initial months. The maximum absolute systematic error in the CZBC hindcasts is about a factor of 2 larger than in the HCM hindcast. In terms of the phase, the CZBC systematic error has relative minima in boreal spring when the HCM systemic errors attain their maxima. The relative maxima in the CZBC systematic errors are during boreal winter. While the phasing of the systematic error is different in the two prediction systems, the forecasts initialized in the first half of the year have larger systematic errors than the forecasts initialized in the second half of the year for both prediction systems.

c. Forecast skill

The 18-month evolution of the SSTA in the NINO3 region for all 60 HCM hindcasts is shown in Fig. 11a. The observed NINO3 SSTA is indicated by the bold solid line. The short-dashed line indicates the evolution of the individual hindcasts. In the same format, the results from the CZBC prediction system are shown in Fig. 11c. Figure 11b shows a smaller sample of hindcasts from the COLA prediction system. The COLA hindcasts include 1982, 1983, 1984, 1986, 1987, 1988, 1989, and 1991, representing 32 cases to compare with the 60 cases in Figs. 11a and 11c.

Focusing on the HCM (Fig. 11a), the hindcasts of the three major warm events (1982–83, 1986–87, and 1991–92) track the observed SSTA fairly well for the first 10–12 months. Compared to the hindcasts of the warm events, the hindcasts of the two cold periods (1984–85, 1988–89) are particularly poor. After the 1991–92 warm event, the HCM hindcasts do not capture the observed SSTA variability. This relatively poor hindcast skill in the mid-1990s is common to many predictions systems. The COLA hindcasts (Fig. 11b) track the observed SSTA somewhat better than the HCM, particularly during the cold events, which is most notable in late 1983 to early 1984 and in 1988–89. On the other hand, the amplitudes of the 1982–83, 1991–92, and perhaps 1986–87 warm events are better simulated by the HCM.

The general character of the CZBC hindcasts is markedly different than the COLA or the HCM hindcasts. The evolution of the individual CZBC hindcasts are considerably smoother and there is a larger degree of consistency between neighboring forecasts. Both of these differences are due to the coupled initialization procedure, which reduces the initial condition “shock” and guarantees that the oceanic and atmospheric states are more internally consistent. The AGCM winds have variability on timescales that are short compared to ENSO and this “noisiness” also contributes to the larger spread in the COLA and HCM hindcasts. While the hindcast skill of the CZBC predictions is higher than either the HCM or the COLA prediction systems, there are hindcasts that are markedly different from the observations, such as those initialized in 1989. It is interesting to note that both the COLA and the HCM prediction systems have similar difficulty simulating the transition from the 1988–89 cold period to near-normal conditions in 1990. The CZBC prediction system also has a tendency to overpredict the amplitude of the warm events, as evidenced by the 1986–87 and 1991–92 warm events.

The SSTA hindcast “skill” shown in Figs. 12a and 12b is consistent with the evolution of the SSTA seen in Figs. 11a and 11c. Figure 12a shows the correlation coefficient and Fig. 12b shows the rmse of the NINO3 hindcast SSTA for the HCM, the CZBC prediction systems, and persistence, respectively. In calculating the skill scores, the systematic errors shown in Figs. 10a and 10b have been removed. Removing the systematic error in the HCM tends to reduce the rmse of the hindcast (between 0.01°C and 0.1°C depending on lead time) but has only a very small affect on the correlation coefficient. (Note that the evolution of the NINO3 temperature shown in Figs. 11a–c includes the systematic error). The skill scores for the HCM (solid curve) are based on all 60 hindcasts shown in Fig. 11a. Similarly, the skill of the CZBC system (dashed curve) and persistence (dot-dot-dashed curve) is based on the same sample of 60 cases.

For all lead times, the CZBC prediction system has a correlation coefficient that is larger than the HCM. Using a correlation of 0.6 as minimum measure for useful forecasts, the HCM is skillful for lead times up to 11 months. Beyond 11 months, the HCM correlation falls off rapidly, whereas the CZBC hindcasts have only a modest reduction in skill. A persistence forecast is more skillful than the HCM for the first five months and more skillful than the CZBC hindcasts for the first three months. Although not shown here, when computing the correlation with the same 32 cases in the overlapping COLA hindcasts, the COLA prediction system is more skillful than the HCM but remains substantially less skillful than the CZBC system. In terms of the rmse, both prediction systems have similar errors up to lead times of 12 months. Beyond 12 months, the HCM has larger rmse.

The spatial structure of the SSTA correlation coefficient for the HCM and for persistence is shown in Figs. 13a–h. For a lead time of three months (Figs. 13a and 13b), a persistence forecast is more skillful particularly in the central Pacific and at subtropical latitudes. By six months (Figs. 13c and 13d), the HCM has larger correlation in the eastern Pacific and persistence has larger correlation in the west-central Pacific. The region of relatively high correlation coefficient for the HCM narrows at nine months and is larger than persistence throughout most of the central and eastern Pacific. At 12 months, persistence has no skill, and while the correlation coefficient is below 0.6 everywhere, the HCM appears to capture a coherent signal in the eastern Pacific.

A potential advantage the HCM has over the CZBC prediction system is that it directly predicts the global circulation and rainfall anomalies in response to the tropical Pacific SSTA. While capturing the extratropical atmospheric response is an area of active research, accurately simulating tropical Pacific rainfall anomalies (or heating anomalies) is a prerequisite to predicting extratropical atmospheric anomalies. With this prerequisite in mind, the precipitation anomaly correlation coefficient in the tropical Pacific for lead times of five to seven months is shown in Fig. 14a. In calculating the correlation for the HCM hindcasts and persistence (Fig. 14b), Microwave Sounding Unit (MSU) rainfall data (Spencer 1993) has been used. Encouragingly, the HCM has correlations greater the 0.4 over a sizeable region in the eastern Pacific. Given the fact that the simulated and observed positive rainfall anomalies are largest in the eastern Pacific (i.e., Figs. 7a and 7b), we expect this region to have the largest correlation. However, between 150°E and 180°, where the largest negative anomalies occur in the observations, the HCM has no skill largely because of an inability to simulate cold events. There is also an indication of a predictable signal over northeast Australia. While persistence appears to have no skill for these lead times, it may be possible to produce an empirical precipitation forecast based on the dynamical SSTA forecast that is as skillful or even better than the HCM precipitation forecast; this issue deserves further study.

5. Hindcast examples

The following examples of the HCM hindcasts are shown to highlight the strengths and weaknesses of this prediction system. The difficulty in predicting the transition from warm to normal or cold conditions is highlighted, as is the problem in forecasting cold events.

The hindcast initialized on 1 January 1982 was particularly successful as can be seen by Figs. 15a,b and Figs. 16a,b. Figures 15a and 15b show the time–longitude cross section along the equator of the hindcast and observed SSTA for the entire 18-month prediction. Similarly, Figs. 16a and 16b show the hindcast and observed precipitation anomaly. The positive SSTA in the east Pacific becomes apparent in the hindcast at approximately the same time as the observed but reaches its peak amplitude somewhat later than the observed. Once the warming begins, the hindcast SSTA, like the observed SSTA, remains positive through the 18th month of the hindcast. The timing of the largest HCM precipitation anomaly in the eastern Pacific (Fig. 16a) is in good agreement with the observations (Fig. 16b); however, the anomaly is about half the magnitude of the observed.

The hindcast initialized six months earlier on 1 July 1981 (Figs. 17a and 17b) fails to forecast the development of the warm ENSO of 1982–83. Initially, both the observed and the hindcast SSTA is slightly below normal. As the observed SSTA warms in March–April 1982, the hindcast also indicates weak warming but fails to intensify as does the observed during the boreal summer of 1982. We have compared the initial thermocline anomaly for the July 1981 and the January 1982 hindcasts and found that the initial downwelling anomaly in the western Pacific in the July 1981 case is considerably weaker than in the January 1982 case, possibly explaining why the July 1981 forecast fails to capture the 1982–83 warm event.

The HCM often has difficulty simulating the transition from warm conditions to either cold or normal conditions. One example is shown in Figs. 18a and 18b. Starting in January 1983, both the HCM and the observations indicate relatively warm conditions that slowly dissipate over the next eight months. By September 1983, the observed SSTA is near normal and the HCM SSTA is about 1°C above normal in the eastern Pacific. However, as the hindcast progresses, the HCM SSTA remains at about 1°C above normal in the eastern Pacific, whereas in the observations, the SSTA has dropped substantially below normal.

Problems hindcasting cold SSTA are not restricted to transitions from warm events. When initialized with relatively cold conditions, the HCM often fails to adequately maintain or amplify the cold anomalies. A typical example of this problem can be seen in Figs. 19a and 19b. In the hindcast during the first four months, there are weak cold anomalies in the far eastern Pacific. After the fourth month, the HCM SSTA in the eastern Pacific becomes slightly positive, whereas the observed SSTA is well below normal. This inability to simulate cold events is consistent with the extended simulation shown in section 3 where cold anomalies were weak and short lived.

6. Summary and concluding remarks

An HCM model was presented that consists of a tropical Pacific ocean model coupled to global spectral atmospheric general circulation model. Typically, the hybrid coupling strategy uses sophisticated OGCMs coupled to simplified or statistical atmospheric models. The approach advanced here reverses which component model contains the more sophisticated dynamics and physics. The motivation for selecting this particular ocean model is based on earlier results with the COLA prediction system, which uses the same AGCM but a more sophisticated OGCM. In the COLA prediction system it was found that the meridional structure of the simulated SSTA was severely truncated in the eastern Pacific compared to the observed SSTA. Moreover, Kirtman et al. (1997) speculated that the erroneously narrow meridional structure potentially limits the predictability of the model. An extended simulation with the HCM indicates that the ocean component used in this HCM model does not suffer from this meridional-scale problem.

A 40-yr simulation with the HCM model produced somewhat realistic and irregular ENSO variability. There were substantial active and inactive periods. A power spectra analysis of the observed and simulated NINO3 SSTA indicated that the HCM variability was weaker and broader than the observed and had too much power in periods on the order of 24 months. The dominant period in the observed record was approximately 42 months and the largest spectral peak in the simulation was at 38 months.

Time–longitude cross sections of the simulated and observed SSTA, zonal wind stress anomaly, and precipitation anomaly indicated primarily a standing ENSO oscillation with largest SSTA in the eastern Pacific. The most outstanding problem in the simulation is the lack of any significant cold events. The largest zonal wind stress anomaly was in the central Pacific primarily to the east of 180° and no eastward propagation was apparent. In the observed data, there is a suggestion of an eastward propagating zonal wind stress anomaly. The precipitation anomaly during warm events is largest in the eastern Pacific, much like the observed anomaly. The observed precipitation anomaly during cold events is largest in the west-central Pacific, and, consistent with the simulated SSTA, the HCM produces no significant negative rainfall anomaly anywhere in the tropical Pacific. This inability to simulate cold events was also apparent in the hindcasts.

An ensemble of 60 hindcast predictions initialized each January, April, July, and October of 1980–94 were analyzed. The initialization strategy for the hindcasts follows that suggested by Kirtman and Schneider (1996) and is also used in the COLA prediction system. The hindcasts were analyzed and compared to the COLA and CZBC prediction systems. In general, the character of the hindcasts is similar to that of the COLA system but substantially different from the CZBC system. The CZBC hindcasts are smoother and there is larger agreement between neighboring hindcasts. On the other hand, the CZBC forecasts tend to overpredict the amplitude of the warm events. As expected from the results of the extended simulation, the HCM fails to predict much of the amplitude of the cold events. There is an indication that the HCM better simulates the amplitude of warm events than the COLA prediction system.

The “skill” of the three prediction systems was discussed primarily in terms of the correlation coefficient. The HCM overall skill at predicting NINO3 SSTA for lead times of up to 11 months is useful, although less than that of the CZBC prediction system. Beyond 11 months, the CZBC system clearly has superior skill. Unlike the CZBC prediction systems, the HCM loses skill rapidly in the first four months, but the correlation remains above 0.6. After the fourth month the correlation levels off until the eleventh month when the skill falls off rapidly. The HCM has greater skill than persistence for all lead times greater than five months.

The hindcast precipitation anomaly skill in the tropical Pacific was also analyzed. For lead times of five to seven months, the HCM produces precipitation anomalies that agree reasonably well with the observed anomalies in the eastern Pacific. For these lead times, a persistence precipitation anomaly forecast has no skill. The fact that the hindcast precipitation anomaly skill was highest in the eastern Pacific is consistent with the extended simulation where it was found that the only significant positive rainfall anomalies were located in the eastern Pacific. There is also an indication of significant skill over northeast Australia.

Several examples of specific hindcasts were shown. These examples highlight two systematic problems with the HCM prediction system. First, the HCM has difficulty in predicting the transition from warm conditions to near-normal or cold conditions. Second, when initialized with cold surface conditions, the HCM fails to maintain or amplify the SSTA as indicated by the observed evolution. The relatively poor simulation of the cold events is likely due to the ocean component in the HCM. DeWitte and Perigaud (1996) have found that the ZC ocean model with observed wind stress forcing does a relatively poor job of simulating cold events because the asymmetry in the parameterization of entrained temperature at 50 m. DeWitte and Perigaud (1996) found that by removing this asymmetry the simulation of cold events improved. However, it remains to be tested how this change would affect the HCM extended simulations and hindcast experiments.

While the skill of the HCM is not as good as either the COLA or the CZBC predictions systems, it shows potential for operational prediction. In fact, a possible approach would be to use both the COLA and HCM prediction systems, combining their forecasts with appropriate weights. For example, the HCM better simulates the amplitude of warm events than the COLA system, so that, for warm events, the HCM forecast would be used. During cold events the COLA system is more skillful and, therefore, would be more heavily weighted.

The success of the CZBC prediction system is primarily due to the coupled initialization procedure. As the present forecast system uses the same ocean model, we expect that it will benefit from comparable improvements with similar forecast initialization procedures. There are other potential improvements to the forecast system. In particular, in uncoupled simulations DeWitt (1996) has found that the relaxed Arakawa–Schubert convection scheme improves the simulation of the global circulation and rainfall anomalies as well as the tropical Pacific wind stress anomalies (Kirtman and DeWitt 1997). It may also be possible to include the ocean model iteration procedure in some statistical way in the prediction experiments. Modifications to the ocean model parameterizations (e.g., DeWitte and Perigaud 1996) also have the potential to improve the coupled simulation. These issues are being considered as future model development.

Acknowledgments

Professor Yunqi Ni (Nanjing University) performed preliminary research in coupling these models and his earlier successes have motivated the continuation of the project. This work has benefitted from discussions with E. Schneider, J. Shukla, and M. Cane. D. Straus provided algorithms to calculate the red noise spectral density. We are also grateful to J. Kinter for editing the manuscript. B. Kirtman is supported under NOAA Grant NA26-GPO149, NA46-GPO217 and NSF Grant ATM-93021353. S. Zebiak’s contribution to this research was supported by NOAA Grant NA56-GPO221 and NSF Grant ATM-9224915.

REFERENCES

  • Barnett, T. P., M. Latif, N. Graham, M. Flugel, S. Pazan, and W. White, 1993: ENSO and ENSO-related predictability. Part I: Prediction of equatorial Pacific sea surface temperature with a hybrid coupled ocean–atmosphere model. J. Climate,6, 1545–1566.

  • Barnston, A. G., and C. F. Ropelewski, 1992: Prediction of ENSO episodes using canonical correlation analysis. J. Climate,5, 1316–1345.

  • Cane, M. A., and S. E. Zebiak, 1985: A theory for El Niño and the Southern Oscillation. Science,228, 1085–1087.

  • ——, and ——, 1987: Predictions of El Niño events using a physical model. Atmospheric and Oceanic Variability, H. Cattle, Ed., Royal Meteorological Society Press, 153–182.

  • ——, ——, and S. C. Dolan, 1986: Experimental forecasts of El Niño. Nature,321, 827–832.

  • Chen, D., S. E. Zebiak, A. J. Busalacchi, and M. A. Cane, 1995: An improved procedure for El Niño forecasting. Science,22, 1699–1702.

  • Davey, M. K., S. Ineson, and M. A. Balmaseda, 1994: Simulation and hindcast of tropical Pacific Ocean interannual variability. Tellus,46A, 433–447.

  • DeWitt, D. G., 1996: The effect of the cumulus convection on the climate of the COLA general circulation model. COLA Tech. Rep. 27, 69 pp. [Available from COLA, 4041 Powder Mill Road, Suite 302, Calverton, MD 20705.].

  • DeWitte, B., and C. Perigaud, 1996: El Niño–La Niña events simulated with Cane and Zebiak’s model and observed with satellite or in situ data. Part II: Model forced with observations. J. Climate,9, 1188–1207.

  • Goldenberg, S. B., and J. J. O’Brien, 1981: Time and space variability of the tropical Pacific wind stress. Mon. Wea. Rev.,109, 1190–1207.

  • Halpert, M. S., and C. F. Ropelewski, 1992: Surface temperature patterns associated with the Southern Oscillation. J. Climate,5, 577–593.

  • Harshvardhan, R. Davis, D. A. Randall, and T. G. Corsetti, 1987: A fast radiation parameterization for general circulation models. J. Geophys. Res.,92, 1009–1016.

  • Huang, B., and E. K. Schneider, 1995: The response of an ocean general circulation model to surface wind stress produced by an atmospheric general circulation model. Mon. Wea. Rev.,123, 3059–3085.

  • ——, and J. Shukla, 1997: An examination of the AGCM simulated surface wind stress and low-level winds over the tropical Pacific Ocean. Mon. Wea. Rev,125, 985–998.

  • Jenkins, G. M., and D. G. Watts, 1968: Spectral Analysis and Its Applications. Holden-Day, 525 pp.

  • Ji, M., A. Kumar, and A. Leetmaa, 1994a: An experimental coupled forecast system at the National Meteorological Center: Some early results. Tellus,46A, 398–418.

  • ——, ——, and ——, 1994b: A multi-season climate forecast system at the National Meteorological Center. Bull. Amer. Meteor. Soc.,75, 569–577.

  • ——, A. Leetmaa, and V. E. Kousky, 1996: Coupled model predictions of ENSO during the 1980s and the 1990s at the National Centers for Environmental Prediction. J. Climate,9, 3105–3120.

  • Kinter, J. L., III, J. Shukla, L. Marx, and E. K. Schneider, 1988: A simulation of winter and summer circulations with the NMC global spectral model. J. Atmos. Sci.,45, 2486–2522.

  • Kirtman, B. P., 1997: Oceanic Rossby wave dynamics and the ENSO period in a coupled model. J. Climate,10, 1690–1704.

  • ——, and E. K. Schneider, 1996: Model based estimates of equatorial Pacific wind stress. J. Climate,9, 1078–1091.

  • ——, and D. G. DeWitt, 1997: Comparison of atmospheric model wind stress with three different convective parameterizations: Sensitivity of tropical Pacific Ocean model simulations. Mon. Wea. Rev.,125, 1231–1250.

  • ——, A. Vernekar, D. DeWitt, and J. Zhou, 1993: Impact of orographic gravity wave drag on extended range forecasts with the COLA-GCM. Atmosfera,6, 3–24.

  • ——, J. Shukla, B. Huang, Z. Zhu, and E. K. Schneider, 1997: Multiseasonal predictions with a coupled tropical ocean–global atmosphere system. Mon. Wea. Rev.,125, 789–808.

  • Kleeman, R., 1993: On the dependence of hindcast skill on ocean thermodynamics in a coupled ocean–atmosphere model. J. Climate,6, 2012–2033.

  • Kuo, H. L., 1965: On the formation and intensification of tropical cyclones through latent heat release by cumulus convection. J. Atmos. Sci.,22, 40–63.

  • Lacis, A. A., and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth’s atmosphere. J. Atmos. Sci.,31, 118–133.

  • Latif, M., and A. Villwock, 1990: Interannual variability as simulated in coupled ocean–atmosphere models. J. Mar. Syst.,1, 51–60.

  • ——, A. Sterl, E. Maier-Reimer, and M. M. Junge, 1993a: Climate variability in a coupled GCM. Part I: The tropical Pacific. J. Climate,6, 5–21.

  • ——, ——, ——, and ——, 1993b: Structure and predictability of the El Niño/Southern Oscillation phenomena in a coupled ocean–atmosphere general circulation model. J. Climate,6, 700–708.

  • ——, T. P. Barnett, M. A. Cane, M. Flugel, N. E. Graham, H. von Storch, J.-S. Xu, and S. E. Zebiak, 1994: A review of ENSO prediction studies. Climate Dyn.,9, 167–179.

  • Leetmaa, A., and M. Ji, 1989: Operational hindcasting of the tropical Pacific. Dyn. Atmos. Oceans,13, 465–490.

  • Mellor, G. L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys.,20, 851–875.

  • Miyakoda, K., and J. Sirutis, 1977: Comparative integrations of global spectral models with various parameterized processes of sub-grid scale vertical transports. Beitr. Phys. Atmos.,50, 445–48.

  • Neelin, J. D., 1990: A hybrid coupled general circulation model for El Niño studies. J. Atmos. Sci.,47, 674–693.

  • Pacanowski, R. C., K. Dixon, and A. Rosati, 1993: The GFDL modular ocean model users guide, version 1.0. GFDL Ocean Group Tech. Rep. 2, 77 pp. [Available from NOAA/GFDL, Princeton University, Princeton, NJ 08542.].

  • Parker, D. E., M. Jackson, and E. B. Horton, 1995: The GISST sea surface temperature and sea ice climatology. Hadley Centre Climate Research Tech. Note 63, 49 pp. [Available from Hadley Centre Meteorological Office, London Road, Bracknell, Berkshire RG12 2SY, United Kingdom.].

  • Penland, C., and T. Magorian, 1993: Prediction of Niño 3 sea surface temperatures using linear inverse modeling. J. Climate,6, 1067–1076.

  • Philander, S. G. H., R. C. Pacanowski, N. C. Lau, and M. J. Nath, 1992: Simulation of ENSO with a global atmospheric GCM coupled to a high-resolution, tropical Pacific Ocean GCM. J. Climate,5, 308–329.

  • Robertson, A. W., C.-C. Ma, M. Ghil, and C. R. Mechoso, 1995: Simulation of the tropical Pacific climate with a coupled ocean–atmosphere general circulation model. Part II: Interannual variability. J. Climate,8, 1199–1216.

  • Reynolds, R. W., 1988: A real time global sea surface temperature analysis. J. Climate,1, 75–86.

  • Schneider, E. K., and J. L. Kinter III, 1994: An examination of internally generated variability in long climate simulations. Climate Dyn.,10, 181–204.

  • ——, Z. Zhu, B. Huang, B. Giese, B. P. Kirtman, and J. Shukla, 1997: ENSO variability in a coupled general circulation model. Mon. Wea. Rev., in press.

  • Spencer, R. W., 1993: Global oceanic precipitation from the MSU during 1979–91 and comparison to other climatologies. J. Climate,6, 1301–1326.

  • Tiedtke, M., 1984: The effect of penetrative cumulus convection on the large-scale flow in a general circulation model. Beitr. Phys. Atmos.,57, 216–239.

  • Trenberth, K. E., W. G. Large, and J. G. Olson, 1990: The mean annual cycle in global ocean wind stress. J. Phys. Oceanogr.,20, 1742–1760.

  • Xue, Y., P. J. Sellers, J. L. Kinter III, and J. Shukla, 1991: A simple biosphere model for global climate studies. J. Climate,4, 345–364.

  • Zebiak, S. E., and M. A. Cane, 1987: A model of El Niño and the Southern Oscillation. Mon. Wea. Rev.,115, 2262–2278.

  • ——, and ——, 1991: Natural climate variability in a coupled model. Workshop on Greenhouse-Gas-Induced Climate Change: A Critical Appraisal of Simulations and Observations, M. E. Schlesinger, Ed., Elsevier, 457–470.

Fig. 1.
Fig. 1.

Temporal correlation between uncoupled ocean model simulated and observed SSTA. In (a) the wind stress anomaly is determined directly from the AGCM and in (b) the AGCM wind stress is empirically derived from the AGCM 850-mb winds. In (c) the wind stress is determined from the FSU analysis. In (d) the wind stress is determined from the ocean model iteration procedure.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 2.
Fig. 2.

NINO3 (5°S–5°N, 150°–90°W) averaged sea surface temperature anomaly for the HCM simulation. In (a) simulation years 1–22 and in (b) simulation years 23–44 are shown.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 3.
Fig. 3.

NINO3 (5°S–5°N, 150°–90°W) averaged sea surface temperature anomaly for the observations. In (a) the SSTA for 1949–69 and in (b) the SSTA 1970–91 is shown.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 4.
Fig. 4.

Power spectra of the (a) observed and (b) simulated NINO3 SSTA. The solid curve shows the power spectra, the long-dashed curve shows the power spectra for red noise, and the short-dashed curve shows the red noise power spectra at the 95% confidence level.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 5.
Fig. 5.

Time–longitude cross section of (a) simulated and (b) observed SSTA along the equator. For the HCM, the simulation years 33–44 are shown, and for the observation, 1982–93 is shown.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 6.
Fig. 6.

Time–longitude cross section of (a) simulated and (b) observed zonal wind stress anomaly along the equator. For the HCM, the simulation years 33–44 are shown, and for the observation, 1982–93 is shown.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 7.
Fig. 7.

Time–longitude cross section of (a) simulated and (b) observed precipitation anomaly along the equator. For the HCM, the simulation years 33–44 are shown, and for the observation, 1982–93 is shown.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 8.
Fig. 8.

SSTA first empirical orthogonal function of the (a) HCM, (b) observations, and (c) COLA prediction system. The explained variance is noted in each panel.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 9.
Fig. 9.

NINO3 SSTA, for the observations (solid curve), the uncoupled ocean model before (dot-dashed curve) and after (short-dashed curve) the iteration procedure.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 10.
Fig. 10.

NINO3 SSTA systematic error in the HCM (top panel) and the CZBC (bottom panel) hindcasts. The thick solid curve in the top panel shows the observed annual cycle scaled by a factor of 3. The systematic error of the January (dot–dashed curve), the April (short-dashed curve), the July (long-dashed curve), and the October (thin solid curve) hindcast are shown separately.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 11.
Fig. 11.

Evolution of the NINO3 SSTA observations (thick solid curve) and the hindcasts (short-dashed curve). The top panel shows the HCM hindcasts, the middle panel shows the COLA hindcasts, and the bottom panel show the CZBC hindcasts.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 12.
Fig. 12.

NINO3 SSTA (a) correlation coefficient and (b) root-mean-square error (rmse). The skill of the HCM is shown in the solid curve and the skill of the COLA system is shown in the long-dashed curve. The skill of the CZBC system is shown in the short-dashed curve, and the skill of persistence is shown in the dot–dot-dashed curve.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 13.
Fig. 13.

SSTA correlation of the HCM hindcasts as a function of latitude, longitude, and time in (a), (c), (e), and (g). SSTA correlation of a persistence hindcast as a function of latitude, longitude, and time in (b), (d), (f), and (h).

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 14.
Fig. 14.

(a) HCM precipitation anomaly correlation coefficient and (b) persistence precipitation anomaly correlation coefficient for lead time of 5–7 months.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 15.
Fig. 15.

Time–longitude cross section along the equator of the (a) January 1982 hindcast and (b) observed SSTA.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 16.
Fig. 16.

Time–longitude cross section along the equator of the (a) January 1982 hindcast and (b) observed precipitation anomaly.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 17.
Fig. 17.

Time–longitude cross section along the equator of the (a) July 1981 hindcast and (b) observed SSTA.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 18.
Fig. 18.

Time–longitude cross section along the equator of the (a) January 1983 hindcast and (b) observed SSTA.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

Fig. 19.
Fig. 19.

Time–longitude cross section along the equator of the (a) January 1984 hindcast and (b) observed SSTA.

Citation: Monthly Weather Review 125, 10; 10.1175/1520-0493(1997)125<2620:ESAPWA>2.0.CO;2

1

The purely statistical methods such as Penland and Magorian (1993) and Barnston and Ropelewski (1992) are not discussed here because they do not explicitly include coupling between the ocean and atmosphere.

Save